QSAR & predictive modelling.
Interpretable machine-learning models for ADMET, target activity and toxicity prediction. We build, benchmark and deploy them with our SIBILA AutoML framework, and expose them as target-specific web servers for diabetes, obesity, anti-aging and natural compounds.
What we work on
Concrete predictive modelling problems we solve for in-house programs and external partners.
- ADMET prediction — absorption, distribution, metabolism, excretion and toxicity models for hit triage.
- Target activity prediction — QSAR models trained on curated datasets, with applicability domain analysis.
- Toxicity and safety — hERG, cytotoxicity and tissue-specific risk models.
- Interpretability — every prediction is delivered with feature attribution (SHAP, LIME, descriptor importance), not as a black box.
- Target-specific servers — public web tools for anti-diabetic, anti-obesity, anti-aging and antioxidant activity.
Tools we use
- SIBILAAutoML platform for interpretable predictive models.
- DIA-DBDiabetes drug prediction by similarity and inverse virtual screening.
- OBE-DBAnti-obesity drug prediction.
- AntiAge-DBNatural cosmetic anti-aging compound prediction (consolidating the former NC-DB resource).
Applications & target areas
Where interpretable QSAR is delivering value for our partners today.
Pharma R&D
Early ADMET and toxicity filtering of compound libraries before in vitro validation.
Metabolic disease
Anti-diabetic and anti-obesity activity prediction for repurposing and natural products.
Cosmetics
Anti-aging ingredient screening through AntiAge-DB and custom QSAR models.
Clinical & environmental
Cardiovascular risk, hospital-readmission and drought-monitoring models built on the same SIBILA stack.
Selected papers
Reference publications underpinning this line.
| Topic | Reference |
|---|---|
| SIBILA — interpretable AutoML platform | 10.3390/ai5040116 |
| DIA-DB — diabetes drug prediction server | 10.1021/acs.jcim.0c00107 |
| OBE-DB — anti-obesity drug prediction | preprint 10.1101/2025.04.10.648110 |
| AntiAge-DB — cosmetic anti-aging predictions | 10.3390/antiox11112268 |
Need an interpretable model for your target, endpoint or dataset?
Prof. Horacio Pérez-Sánchez · hperez@ucam.edu